中文 EN

Bubble's Brain - 2025-12-04

Daily Brief

AI News 2025-12-04

AI Daily Brief

Summary

Google launched Workspace Studio, a no-code AI agent builder for enterprises to automate workflows.
Xiaomi disclosed an AI roadmap focused on integrating large models with physical products and plans a 100B+ parameter multimodal model.
Kuaishou released Kling 2.6 for audio-video co-generation, and OpenAI signed an agreement to acquire Neptune to strengthen training capabilities.

Today’s AI News

  1. Google releases Workspace Studio: a no-code AI agent builder for enterprises: On Dec 3, Google officially launched Google Workspace Studio, a “no-code” AI agent builder for enterprise users. Powered by Gemini 3 multimodal capabilities, it lets non-technical employees describe requirements in natural language and create custom AI agents in minutes—for workflows like email triage and report generation. Workspace Studio offers a no-code UI and templates and supports end-to-end automation across native apps like Gmail and Docs plus third-party platforms like Asana and Salesforce, aiming to break information silos and improve collaboration. Rollout will be phased for Business and Enterprise plans.

  2. Xiaomi reveals AI roadmap: deep integration of large models with physical scenarios: Xiaomi president Lu Weibing outlined Xiaomi’s AI strategy, emphasizing “deep integration of large models with physical scenarios” and embedding AI across phones, cars, and smart home hardware and services. He confirmed Luo Fuli joined to lead the Xiaomi MiMo model team, reporting directly to Lei Jun, with a goal to release a 100B+ parameter multimodal model in 2025. Lu said Xiaomi’s 800M+ consumer IoT devices form a unique data advantage, and the company plans to publish an AI technology whitepaper in Q1 next year.

  3. Ant’s Lingguang AI assistant adds “flash games” creation in 30 seconds: Ant Group’s general AI assistant Lingguang shipped a major update adding “flash games.” Users with no programming background can create and edit mini-games via natural language—changing characters, adjusting rules, and more—in as fast as 30 seconds. It’s an application scenario expansion of Lingguang’s “flash apps” concept, aimed at lowering the barrier for rapid, spontaneous creation. As of Dec 2, users reportedly created 3.3M flash apps on the platform.

  4. WordPress introduces Telex, an experimental AI tool to generate Gutenberg blocks: WordPress launched Telex, an experimental AI developer tool to generate Gutenberg blocks. Though still in testing, it has been used in production sites, enabling features that previously required expensive custom development—like price comparison widgets, real-time store hours, and partner-logo carousels—to be generated in seconds. A WordPress co-founder said it dramatically reduces development time and cost, and the project plans to introduce AI model evaluation standards in 2026 to improve platform intelligence.

  5. Microsoft denies reports of lowering AI growth targets due to sales underperformance: Microsoft denied media reports claiming it reduced AI business growth expectations because some sales teams failed to hit targets. The company said the report conflated individual sales quotas with company-level growth goals and that it has not changed overall growth targets. Despite the clarification, the story triggered investor concerns and briefly pushed the stock down over 2%.

  6. Meta hires Apple design veteran Alan Dye for its AI glasses push: To accelerate its AI glasses project, Meta hired senior designer Alan Dye from Apple, where he led major UI and visual design efforts. The hire is viewed as a strategic upgrade in Meta’s hardware–software integration. Mark Zuckerberg believes wearable AI devices will eventually replace smartphones, and Meta aims to improve AI glasses design and UX while enabling seamless voice and gesture interactions.

  7. BAAI releases Emu3.5 multimodal world model that tokenizes everything into one sequence: The Beijing Academy of Artificial Intelligence (BAAI) released a new multimodal world model Emu3.5. Its core breakthrough encodes images, text, and video into the same token sequence, focusing on predicting the “next world state.” By learning cross-modal causal relationships and physical commonsense, Emu3.5 aims to move from “generating pixels” to “simulating the world,” with potential applications in embodied robotics and autonomous driving simulation.

  8. Anthropic begins IPO preparations and seeks a valuation above $300B: Anthropic has reportedly started preparations for an IPO, hiring long-time partner law firm Wilson Sonsini, and could list as early as 2026. The company is also said to be pursuing new financing at a valuation potentially exceeding $300B, far above the $183B valuation disclosed last September. The move reflects a broader trend of AI leaders exploring public markets; competitor OpenAI is also considering an IPO.

  9. Li Auto launches its first AI accessory: Livis AI glasses: Li Auto released its first AI accessory, the Li Auto AI Glasses Livis, starting at ¥1,999. Co-developed with ZEISS, it runs Li Auto’s MindGPT-4o multimodal model and Livis OS to extend AI capabilities beyond the car. The glasses can link with Li Auto vehicles to control functions via voice commands or view HUD information. Hardware highlights include a 36g frame, IP54 dust/water resistance, a 12MP camera, and up to 18.8 hours of daily battery life.

  10. Kuaishou’s Kling AI releases Kling 2.6 for synchronized audio-video generation: Kuaishou launched Kling 2.6, its first “audio-video co-generation” model. Kling 2.6 can generate visuals, natural speech, sound effects, and ambient audio in one pass, supporting both “text-to-audio-video” and “image-to-audio-video” creation. Use cases span monologues, narration, multi-person dialogue, and music performances—aiming to make video creation faster and easier.

  11. Doubao phone assistant responds to WeChat account issues and “high-risk permission” concerns: After users reported WeChat account anomalies and concerns over “high-risk permissions,” the Doubao phone assistant issued a statement on Dec 2. To address some Nubia phone users experiencing WeChat bans when the assistant operated WeChat, Doubao urgently removed the relevant feature and said affected accounts are being gradually unbanned. On privacy accusations, Doubao said the permission is technically necessary, requires explicit user authorization and prompts, and promised it does not store any user screen content in the cloud.

  12. OpenAI signs agreement to acquire Neptune to strengthen model training tooling: OpenAI announced it signed an agreement to acquire startup Neptune, which builds monitoring and debugging tools to improve training efficiency and accuracy. Neptune previously worked with OpenAI on metrics dashboards. After the acquisition, Neptune plans to wind down external services over the coming months and integrate its tooling into OpenAI’s training stack. Neptune has reportedly raised $18M+ from investors.

Comments